Re: [PATCH v5 7/8] multifd: Add multifd-zstd-level parameter

[Dr. David Alan Gilbert]

* Markus Armbruster (address@hidden) wrote:
> Juan Quintela <address@hidden> writes:
> 
> > Markus Armbruster <address@hidden> wrote:
> >> "Dr. David Alan Gilbert" <address@hidden> writes:
> >>
> >>> * Juan Quintela (address@hidden) wrote:
> >>>> Signed-off-by: Juan Quintela <address@hidden>
> >>>> ---
> >>>>  migration/migration.c | 15 +++++++++++++++
> >>>>  monitor/hmp-cmds.c    |  4 ++++
> >>>>  qapi/migration.json   | 29 ++++++++++++++++++++++++++---
> >>>>  3 files changed, 45 insertions(+), 3 deletions(-)
> >>>> 
> >>>> diff --git a/migration/migration.c b/migration/migration.c
> >>>> index 3b081e8147..b690500545 100644
> >>>> --- a/migration/migration.c
> >>>> +++ b/migration/migration.c
> >>>> @@ -91,6 +91,8 @@
> >>>>  #define DEFAULT_MIGRATE_MULTIFD_METHOD MULTIFD_METHOD_NONE
> >>>>  /*0: means nocompress, 1: best speed, ... 9: best compress ratio */
> >>>>  #define DEFAULT_MIGRATE_MULTIFD_ZLIB_LEVEL 1
> >>>> +/* 0: means nocompress, 1: best speed, ... 20: best compress ratio */
> >>>> +#define DEFAULT_MIGRATE_MULTIFD_ZSTD_LEVEL 1
> >>>>  
> >>>>  /* Background transfer rate for postcopy, 0 means unlimited, note
> >>>>   * that page requests can still exceed this limit.
> >>>> @@ -805,6 +807,8 @@ MigrationParameters 
> >>>> *qmp_query_migrate_parameters(Error **errp)
> >>>>      params->multifd_method = s->parameters.multifd_method;
> >>>>      params->has_multifd_zlib_level = true;
> >>>>      params->multifd_zlib_level = s->parameters.multifd_zlib_level;
> >>>> +    params->has_multifd_zstd_level = true;
> >>>> +    params->multifd_zstd_level = s->parameters.multifd_zstd_level;
> >>>
> >>> Do we really want different 'multifd_...._level's or just one
> >>> 'multifd_compress_level' - or even just reuse the existing
> >>> 'compress-level' parameter.
> >>
> >> compress-level,
> >
> > possible values: 1 to 9 deprecated
> >
> >> multifd-zlib-level
> >
> > possible values: 1 to 9, default 1
> > (zlib default is -1, let the lib decide)
> >
> > , and multifd-zstd-level apply
> >
> > possible values: 1 to 19, default 1
> > (zstd default is 3)
> >
> >> "normal" live migration compression, multifd zlib live migration
> >> compression, and multifd zstd live migration compression, respectively.
> >>
> >> Any live migration can only use one of the three compressions.
> >>
> >> Correct?
> >
> > Yeap.
> >
> >>> The only tricky thing about combining them is how to handle
> >>> the difference in allowed ranges;  When would the right time be
> >>> to check it?
> >>>
> >>> Markus/Eric: Any idea?
> >>
> >> To have an informed opinion, I'd have to dig through the migration
> >> code.
> >
> > Problem is _how_ to setup them.  if we setup zstd compression method,
> > put the value at 19, and then setup zlib compression method, what should
> > we do?
> >
> > Truncate to 9?
> > Give one error?
> > Don't allow the zlib setup?
> >
> > Too complicated.
> 
> The interface pretends the parameters are all independent: you get to
> set them one by one.
> 
> They are in fact not independent, and this now leads to difficulties.
> 
> To avoid them, the interface should let you specify a desired
> configuration all at once, and its implementation should then do what it
> takes to get from here to there.
> 
> Example: current state is multifd compression method "zstd", compression
> level is 19.  User wants to switch to method "zlib" level 9 the obvious
> way
> 
>     With old interface:
>         step 1: set method to "zlib"
>         step 2: set level to 9
>     Problem: after step 1, we have method "zlib" with invalid level 19.
>     Workaround: swap the steps.
> 
>     Note that the workaround only works because the set of levels both
>     methods support is non-empty.  We might still come up with more
>     complicated parameter combinations where that is not the case.
> 
>     With new interface:
>         set compression to "zlib" level 9
> 
> The new interface require us to specify a QAPI type capable of holding
> the complete migration configuration.
> 
> The stupid way is to throw all migration parameters into a struct, and
> make the ones optional that apply only when others have certain values.
> Thus, the "applies only when" bits are semantical, documented in
> comments, and enforced by C code.

I realised we already have that stupid struct!  It's just
MigrationParameters - it has all the parameters as optional values,
and QMP's MigrateSetParameters already takes that - so you can already
provide both the type and the level at the same time; although there's
no semantic correlation between them.

Dave

> With a bit more care, we can make "applies only when" syntactical
> instead.  Examples:
> 
>     @max-bandwidth and @downtime-limit always apply.  They go straight
>     into the struct.
> 
>     @tls-creds, @tls-hostname, @tls-authz apply only when TLS is enabled
>     by setting @tls-creds.  Have an optional member @tls, which is a
>     struct with mandatory member @creds, optional members @hostname,
>     @authz.
> 
>     @multifd-zlib-level applies when @multifd-method is "zlib", and
>     @multifd-zstd-level applies when it's "zstd".  Have a union
>     @multifd-compression, cases "none", "zlib" and "zstd", where each
>     case's members are the parameters applying in that case.
> 
> Please note the purpose of these examples is to show how things can be
> done in the schema.  I'm not trying to tell you how these specific
> things *should* be done.
> 
> The resulting type is perfectly suited as return value of a query
> command.  It's awkward as argument of a "specify desired configuration"
> command, because it requires the user to specify *complete*
> configuration.  If we want to support omitting the parts of it we don't
> want to change, we have to make more members optional.  Imprecise for
> the query, where we now have to specify "always present" in comments.
> Usually less bad than duplicating a complex type.
> 
> >> Documentation of admissible range will become a bit awkward, too.
> >>
> >> Too many migration parameters...
> >
> > Sure, but the other option is taking a value and live with it.
> > I am all for leaving the library default and call it a day.
> >
> > Later, Juan.
> 
> Hope this helps some.
> 
> 
--
Dr. David Alan Gilbert / address@hidden / Manchester, UK